Related Experiment Video
Updated: May 28, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
PGT-Net: A Physics-Guided Transformer-CNN Hybrid Network for Low-Light Image Enhancement and Object Detection in
Bin Chen1,2, Jian Qiao3, Baowei Li2
1School of Computer and Artificial Intelligence, Zhengzhou University, Zhengzhou 450002, China.
Journal of Imaging
|May 26, 2026
Summary
This study introduces a new Physically Guided Transformer-CNN Hybrid Network (PGT-Net) to improve low-light image enhancement and object detection for autonomous driving. PGT-Net enhances image quality and detection accuracy in challenging traffic conditions.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Image Processing
Background:
- Low-light conditions degrade image quality, impacting autonomous driving object detection reliability.
- Current deep learning methods lack physical interpretability and robustness in varying traffic scenarios.
Purpose of the Study:
- To develop an end-to-end hybrid network for joint low-light image enhancement and object detection.
- To improve the robustness and interpretability of visual perception systems in adverse lighting.
Main Methods:
- Proposed a Physically Guided Transformer-CNN Hybrid Network (PGT-Net) integrating a physical model with deep learning.
- Utilized a learnable physical guidance branch for atmospheric illumination and transmittance estimation.
- Employed a dual-branch (CNN and Transformer) enhancement backbone with adaptive feature fusion.
Main Results:
- PGT-Net significantly outperforms existing methods in low-light image enhancement (PSNR/SSIM) and object detection (mAP).
- The network maintains high inference efficiency.
- Demonstrated superior performance on public datasets like ExDark and BDD100K-night.
Conclusions:
- PGT-Net offers an interpretable and high-performance solution for visual perception under adverse lighting.
- The physically guided approach enhances robustness in complex traffic scenarios.
- This research holds significant theoretical and practical value for autonomous systems.
